Weights Initialization tab (Network Definition dialog box)

Top  Previous  Next

 

Use this tab to specify options for the initialization of neural connection weights.

 

Options

 

Seed Generator Once

Check this box if you want to reuse the initial randomization seeds every time initialization of the neural connection weights takes place. When this box is selected, neural connection weights initialize to the same values.

 

Use Random Seeds

Check this box to specifies the generation of random seeds every time initialization of the neural connection weights takes place. When this box is selected, neural connection weights initialize to distinct values.

 

Populate Seeds

Check this box to allow or inhibit copying the current layer's seed into the other layers when carrying out the << Populate and Populate >> commands.

 

Specifics

 

Layer

Specifies the currently selected layer. Select from one of the listed layers to modify its initialization settings.

 

Seed

Displays the seed used to generate the initial neural connection weights for this layer. You can type a seed of your preference if the Use Random Seeds box is clear.

 

Minimum

Type in this box the minimum value that can be generated to initialize the neural connection weights.

 

Maximum

Type in this box the maximum value that can be generated to initialize the neural connection weights.

 

NoteThe Minimum value must be less than or equal to the specified Maximum value.

 

Generation Method

 

Range

Select this option to specify generation of random values within the range specified by the Minimum and Maximum boxes.

 

Bipolar

Select this option to specify that the generated values are one of the values specified in the Minimum and Maximum boxes.

 

<< Populate

Copies the current layer initialization settings to all previous layers.

 

Populate >>

Copies the current layer initialization settings to all following layers.

 

NoteSelect the Populate Seeds check box if you want the layer's initialization seed copied as well to the other layers.

 

Sponsored

Try Predictive Systems Lab

Interactive Demo →

Need Help?
Contact support